Feedback analytics for indication search and user profiling

MedCurrent Corporation是一家快速发展的多伦多公司，开发健康IT临床决策支持软件（CDS）。该公司面临的一个主要问题是适应症搜索的性能，这是推荐适当医疗程序的关键步骤。在当前系统中，医生通过自由文本输入医学适应症（例如头痛），并且系统提示匹配适应症的列表。如果找不到匹配项，则输入的文本将保存到系统中。随后，将手动分析所有不匹配的文本，以将其添加到预定义的指示数据库中或丢弃。这一过程对于提高指示搜索的命中率至关重要。然而，这是非常昂贵和低效的。当前系统的另一个问题是没有内置的学习过程，使得搜索和推荐可以从先前的结果中受益。在这个项目中，我们建议应用数据分析和信息检索技术来改善适应症搜索以及整体系统的有效性。我们将使用现有的领域语料库和系统收集的查询日志来构建特定领域的词库，以增加找到匹配指示的机会。对于不匹配的自由文本适应症，我们将应用适当的信息检索机制，使用多个数据源找到最匹配的现有适应症，以减轻未来的人工检查过程。我们还将使用收集到的日志数据来构建用户配置文件，以个性化各个用户的指示搜索和推荐过程，并为客户端生成分析报告（例如，医院），识别各种使用模式。拟议的项目将通过改善决策支持使加拿大卫生部门受益 这是一个订购医疗程序的过程，并向医院提供有关医疗从业人员如何遵守规则和指南的见解。它还将帮助MedCurrent优化其当前的临床支持软件，从而为客户提供更好的体验。

MedCurrent Corporation is a rapidly growing Toronto-based company that develops Health IT Clinical Decision Support Software (CDS). One major problem the company is facing is the performance of the indication search, which is a crucial step for recommending proper medical procedures. In the current system, a physician enters medical indications (e.g. headache) via free-text, and the system prompts with a list of matching indications. If no matching can be found, the entered texts will be saved into the system. Later all the unmatched texts will be analyzed manually to be either added into the pre-defined indication database or discarded. This process is essential to improve the hit ratio of the indication search. However, it is very costly and inefficient. Another problem of the current system is that there is no built-in learning process so that the searching and recommendation can benefit from previous results. In this project, we propose to apply the data analytics and information retrieval techniques to improve the indication search as well as the overall system effectiveness. We would use existing domain corpora and the query logs the system has collected to build a domain-specific thesaurus to increase the chance of finding the matched indications. For the unmatched free-text indications, we would apply proper information retrieval mechanisms to find the best matching existing indications using multiple data sources to alleviate the future manual inspection process. We would also use the collected log data to build user profiles to personalize the indication search and recommendation process for individual users and to generate an analytic report for the client (e.g., hospital), identifying various usage patterns. The proposed project will benefit Canada's health sector by improving the decision support process for ordering medical procedures and giving insights to hospitals on how the rules and guidelines are being followed by the medical practitioners. It would also help MedCurrent optimize their current clinical support software and consequently provide improved experience for their customers.