Semantics based text matching for unstructured text documents

在我们与行业合作伙伴的早期合作中，我们有效地开发了向求职者提供高质量推荐的方法。我们建立了大量的方法，从单词和概念的句法和语义分布到更复杂的语言建模技术，以及通过单词和段落和文档模板对招聘信息的连续表示进行无监督学习。然而，一旦被纳入BCJobs并被采用，人们注意到，更高质量的建议并不一定会导致更多的工作申请。基于这一发现，在这个后续项目中，我们的主要目标将是从交互日志中了解用户行为，以解决以下问题：“影响用户与BCJobs平台互动的参数或条件是什么？”目的是发现为什么用户在第一次查看招聘信息后决定不再继续使用BCJobs。这也要求我们了解是什么鼓励了那些决定参与并最终申请给定工作的用户。本研究采用的主要研究方法是随机对照研究，以确定各种因素对使用寿命的影响。研究工作和所需的开发将与工业合作伙伴BC Jobs Online Inc.密切合作。

In our earlier engagement with the industrial partner, we effectively developed methods that produced qualityrecommendations to job seekers. We built a wide host of methods, ranging from syntactical and statisticaldistribution of words and concepts to more complex language modeling techniques, as well as the use ofunsupervised learning of continuous representations of job postings through word and paragraph and documentembeddings, were used. However, once integrated and employed on BCJobs, it was observed that higherquality recommendations do not necessarily result in more job applications. Based on this finding, in thisfollow-up project, our primary objective will be on understanding user behavior from interaction logs toaddress the following question: 'what are the parameters or conditions that impact user engagement with theBCJobs platform?' with the aim of discovering why users decide not to further engage with BCJobs after theirfirst job posting view. This will also require us to understand what encouraged those users who did, in fact,decide to engage and eventually applied for a given job. The primary research methods used for this projectwill be based on performing randomized controlled studies to determine the impact of various factors on userengagement. The research work and required development will be carried out in close concert with theindustrial partner BC Jobs Online Inc.