Kaoru Shinkawa

Evaluation of real-time captioning by machine recognition with human support

Verbal meetings are important at work, but employees who are deaf or hard of hearing (DHH) find it difficult to participate. Manual real-time captioning is a solution, but professional stenographers are too expensive for routine use. We are exploring the possibilities of real-time captioning that combines Automated Speech Recognition (ASR) and human capabilities, which can dramatically decrease these costs and thus improve the lives of DHH employees. We developed a flexible ASR-based real-time captioning tool that can be used by non-expert captioners to correct the recognized text in practical workplace situations. In this paper, we will report on our early results, focusing on accuracy and latency.

Towards Extracting Recruiters’ Tacit Knowledge Based on Interactions with a Job Matching System

Finding good job matches for elderly workers is becoming a big challenge for aging society. To secure the labor force population, it is necessary to improve the employment rates of the elderly workers and utilize their accumulated knowledge and skills. Matching relies on each recruiter’s tacit knowledge of what we assume as word association information in matching experts’ mind based on their experience. In order to conduct effective job matching for elderly workers, retrieving word associations from the matching experts’ tacit knowledge is necessary for the purpose of generating domain-specific ontology. In this paper, we propose an interactive job matching system that collects recruiters’ interaction data to find word association specific to job matching for elderly workers. Our system is designed for recruitment operations as to acquire the real interaction data with real job opportunity. Our experimental results indicate that the interaction data are effective for identifying word associations that can be used to extract the recruiters’ tacit knowledge.

Interactive Searching Interface for Job Matching of Elderly Workers

In the aging Japanese society, most elderly people still have enough energy to work and have the potential to become essential labor forces. A job matching method that can allocate their unique abilities is required. However, the current job matching relies on each recruiter’s tacit knowledge, and the recruiter assigns only specific candidates to work profiles. In this paper, we propose an interactive job matching system that can reflect the recruiter’s tacit knowledge and help search for diverse elderly workers for each work profile. The results indicate that interactions of the proposed system can improve the matching diversity by retrieving recruiters’ tacit knowledge. When there is a work profile in which it is difficult to extract appropriate keywords, our interactions become most effective.

Online Learning for Long-Query Reduction in Interactive Search for Experienced Workers

For domain specific document searches like job matching, long queries are often given as a detailed information of targets. Previous studies found that higher quality results can be obtained by searching with an optimal subset of words excerpted from a long query. To excerpt the optimal subset of words, query reduction using machine learning techniques has been studied. Supervised learning requires training data with annotation, which is especially difficult for in-domain data because of its specific terminology. In this study, we propose a model that integrates machine learning techniques and manual processing for long-query reduction. We integrated our model into a job matching system that collects manual “interactions” and used them as training data to learn query reduction. Furthermore, we evaluated our model with actual job offerings and expert profile data obtained from a recruitment agency. We found that our proposed model outperformed the baseline in precision, recall, and F-measure. The result suggests that our model could be used for query reduction of interactive search systems of specific domain data.