[T-VCG 2021] Interactive Visual Exploration of Longitudinal Historical Career Mobility Data

[uhhyunjoo]

link
paper	A survey of heterogeneous information network analysis

[uhhyunjoo]

Abstract

• 질적인 historical 데이터셋의 이용가능성이 커짐에 따라, social science 의 여러 분야에 대한 새로운 연구 기회들이 생겨나고 있다.
• 본 논문에서는 새로운 데이터셋인 CGED-Q (China Government Employee Database-Qing) 의 constructors 와 closely 하게 협력했다.
• 해당 데이터셋은, 1760년부터 1912년까지의 중국 Qing 관료제에서의 34만 명 이상의 정부 관리들의 career trajectories (경력 궤적)을 기록했다... (뭐 뭐임 무섭다)
• 본 연구진들은 이 데이터를, 역사적 관점에서 career mobility 를 연구하기 위해, 그리고 social mobility 와 inequality 를 이해하기 위해 사용했다.
• 그러나, 현존하는 통계적인 접근 방식은 이 historical 데이터셋의 fine-grained attributes, long time span 을 이용하여 careere mobility 를 분석하는데 부적절했다.
• 왜냐하면 이들은 hypothesis-driven (가정 주도) 이고, substantial 한 노력을 필요로 하기 때문이다.
• 본 논문에서는, 전문가들이 historical career data 를 이요해서 exploring, understanding, reasoning 을 하는 것을 돕기 위한 an interactive visual analytics system 인, CareerLens 를 제안한다.
• CareerLens 를 사용하면, 전문가들은 mobility patterns 를 3 가지 디테일 레벨로 조사할 수 있고, 각 개인에 대한 social relationships 를 micro-level 로 드러낼 수 있다.
• 본 연구진은 CareerLens 의 효과와 유용함을 2 가지 케이스 스터디를 통해 설명했다. 그리고 domain experts 와의 follow-up interview 를 통해, encouraging feedback 을 받았다.

[uhhyunjoo]

Introduction

• However, to the best of our knowledge, few studies from the visualization community have explored career mobility.

• Previous works that studied career data focused mainly on social network extraction [8], [9], similarity analysis [10], [11], [12], and group-level event sequence summarization [13], [14].

• They have not targeted career mobility analysis, in which overall mobility trends, along with the career mobility of the groups and individuals of interest need to be examined.

• Developing a visualization system for supporting such analysis faces several challenges.

• First, visualizing a large volume of longitudinal career data with a complex data structure is challenging. Depicting the temporal evolution of mobility patterns in such a longitudinal dataset with multiple attributes will incur severe scalability problems given the long time span.

• Second, extracting and highlighting social groups and social relationships from this large dataset is non-trivial. Existing sociological methods for extracting social relationships and latent groups based on similarities [15], [16], [17] require specialized expertise and are ad-hoc. Highlighting the groups and social relationships of interest in their overall context without overwhelming the user with detail heightens the challenge.

• Third, supporting multi-level mobility analysis and reasoning is difficult. Historians and sociologists have difficulty efficiently exploring overall mobility alongside that of groups and individuals. Summarizing the statistics of groups and individuals to identify targets for further investigation and reasoning requires substantial effort.

• To address these challenges, we present CareerLens, a visual analytics system which enables experts to generate and verify hypotheses, and explore and reason about insightful patterns of historical career mobility at three levels-of-detail (LODs).

• At the overall level, the system summarizes the longterm evolution of mobility, allowing experts to explore and compare salient features of groups at different periods.

• At the group level, CareerLens provides recommendations for latent social groups obtained from a group detection algorithm.

• At the individual level, filtering influential persons is also supported.

• The groups and individuals of interest with their social relationships can be further highlighted as group subflows or career threads using our novel flow design.

• It utilizes a multi-scale approach to clearly show the position and proportion of people of interest in their overall population context.

• Moreover, CareerLens is a well-coordinated system which enables experts to explore career mobility effectively.

• We evaluated the effectiveness and usability of the system through two case studies, a longitudinal study, and expert interviews which received positive feedback.