This Small Business Innovation Research (SBIR) Phase II project aims to improve the quality of on-demand job matching by applying data mining and machine learning techniques to natural language descriptions of job requests, worker reviews, and transaction history. The project will enable lasting job matches by predicting the needs, preferences and constraints of workers and human resource managers. Currently available methods of job matching rely primarily on keyword search, corporate personality assessment tests, or fixed ontologies. Such systems lack comprehensive learning and therefore have difficulty matching workers with jobs. This project approaches job matching with a bias-free learning model that learns from hiring successes, trains on real-world data, and adapts to new job verticals.
The broader/commercial impact of the project is a matching technology that optimizes workers' and employers' strengths, discovering matching opportunities overlooked by traditional search technologies. Online reputation-building through performance reviews can improve workers' ability to market themselves. The global matching technology permits nearly every skill to become marketable by matching workers with all features from every available job request. Natural language processing techniques, developed in the course of this project, have the potential to broaden the appeal of cell phone text-messaging as a comprehensive job-searching tool. Furthermore, the contextual approach to learning about workers and employers enables trends to be identified among users, and has far-reaching commercial implications in fields as diverse as medical research and e-commerce.
Research and Development undertaken during this Small Business Innovation Research Phase II project has focused on developing a set of Natural Language Processing (NLP) algorithms for analyzing patient-to-patient conversations taking place on the Internet. The algorithms developed under this grant serve as the foundation for PatientScope™, a software that helps the healthcare industry gain in-depth understanding of the living experience of patients worldwide. Billions of conversations between patients take place on social networks, forums, and blogs. PatientScope aims to understand and quantify these patient experiences as they are reported - automatically, in-depth, and on the Internet scale. Capable of answering a wide range of patient-centric questions, the depth and precision of PatientScope analysis far surpass the capabilities of social listening tools used by the healthcare industry today. In contrast to competing software solutions, PatientScope avoids reducing the rich, multifaceted patient experiences to shallow sentiment scores or volumetric counts of product names. Instead, PatientScope looks for and extracts finer-grained information such as switching between treatments, concerns about drug safety, and comparative treatment efficacy. As a result, the voice of the patient can finally be heard, helping healthcare companies to address patient needs more promptly and effectively than ever before. PatientScope demonstrates that exposing the right patient conversations to enterprise stakeholders and patients themselves carries a great value. The robustness of PatientScope insights suggests that the voice of the patient stands to become the leading determinant of essential healthcare knowledge, from monitoring treatment performance and product defects to aligning pricing and sales strategy with the perceived advantages over competing medications. Three main technical goals were achieved during this Phase II project: Identify and acquire very large sets of healthcare-related text content Scale and adapt NLP techniques to process very noisy healthcare datasets Perform analytics that address both coarse- and fine-grained information needs PatientScope addresses two key needs of healthcare industry stakeholders: Adverse Event Detection: trained to automatically identify adverse events reported across the Internet, PatientScope reduces (and in some cases eliminates) the costly process of manually monitoring safety events across social media. Shown to be highly accurate and cost-effective, PatientScope Safety Monitoring service has been used by some of the world's top pharmaceutical companies. Market Research: PatientScope provides marketing teams within healthcare organizations with the ability to learn from patient chatter without unnecessarily increasing regulatory obligation. Pharmaceutical companies use PatientScope to perform targeted analysis of treatment performance (e.g., effectiveness, risk areas) and public perception (e.g., expense, accessibility), pinpointing opportunities in a transparent and cost-effective fashion, and facilitating quick, data-driven response. The three advantages of using PatientScope are: Broader Coverage: PatientScope covers over 18,500 healthcare products, including prescription and over-the-counter medications, as well as supplements and vitamins. Higher Accuracy: PatientScope removes the majority of Internet SPAM and irrelevant chatter (i.e., posts unrelated to personal healthcare experiences). Elimination of irrelevant posts ensures high quality insights, and serves as a key differentiator vis-a-vis alternative methods for monitoring patient chatter. Deeper Insight: Monitoring concepts - instead of words - allows PatientScope to deliver clear, differentiated data, identifying patients that are diagnosed with a disease, suffer from a symptom, or are prescribed a treatment. Competing solutions monitor words: analyzing a sentence like I tried taking Tylenol for my back pain using traditional text analytics can identify the words (Tylenol, pain) but not the context (What were the patient's symptoms? What medication did they take to cure these symptoms? Was the treatment effective?). As a result, clients using PatientScope can get robust, up to date answers to questions that previously necessitated a small, biased, and expensive surveys. In summary, PatientScope facilitates an unparalleled structuring of patient-contributed information, and frees researchers and patients alike to pose deep questions to the wider public regarding medication efficacy, tolerability, and preference. Continuing this work will enable patient-contributed knowledge to become the dominant method of learning about and finding solutions for health problems facing patients worldwide.
