Yu Chi

Deep keyphrase generation

Keyphrase provides highly-summative information that can be effectively used for understanding, organizing and retrieving text content. Though previous studies have provided many workable solutions for automated keyphrase extraction, they commonly divided the to-be-summarized content into multiple text chunks, then ranked and selected the most meaningful ones. These approaches could neither identify keyphrases that do not appear in the text, nor capture the real semantic meaning behind the text. We propose a generative model for keyphrase prediction with an encoder-decoder framework, which can effectively overcome the above drawbacks. We name it as deep keyphrase generation since it attempts to capture the deep semantic meaning of the content with a deep learning method. Empirical analysis on six datasets demonstrates that our proposed model not only achieves a significant performance boost on extracting keyphrases that appear in the source text, but also can generate absent keyphrases based on the semantic meaning of the text. Code and dataset are available at this https URL.

Understanding and modeling behavior patterns in cross‐device web search

Recent studies have witnessed an increasing popularity of cross‐device web search, in which a user resumes a previously started search task from one device to a later session on another. The complexity of this novel search mode, mainly due to the involvement of interruptions and multiple devices, makes users conduct new types of search behaviors and adopt novel search patterns. Though past research has studied this new search mode, there still lacks a sufficient understanding of cross‐device search behaviors. To better understand users behavioral patterns, we adopted hidden Markov model (HMM) to model observed cross‐device search behaviors and their underlying search pat‐ terns under the same framework. HMM is a widely adopted machine learning technique with the main assumption that observations (e.g., observed search behaviors) are driven by hidden variables (e.g., hidden behavioral patterns). Utilizing the data collected from a user study consisting of two types of cross‐session search conditions – mobile‐to‐desktop (M‐D) and desktop‐to‐desktop (D‐D) – we demonstrate the validity of using HMM for modeling and identifying hidden behavioral patterns, and based on the identified patterns, we discover that re‐finding is a common pattern at the beginning of the continued search session, and querying, exploitation and exploration contribute to the dominated behaviors in later search stages. Our results also show clear device effects on cross‐device search (M‐D) vs cross‐session search (D‐D). All these evidences confirm that better support mechanisms are in critical need for a cross‐device search and the HMM‐ based modeling can help.

“It lives all around us”: Aspects of data literacy in teen's lives: “It Lives All Around Us”: Aspects of Data Literacy in Teen's Lives

In this paper, we examine young peoples data literacy in terms of their awareness of data and the rhetoric that surrounds it, as well as their knowledge of data flows. This is the first phase of the Exploring Data Worlds at the Public Library research study research study, a two‐year research study that explores the ways that libraries can address data literacy programming by helping teens understand, create and manage the digital traces of their data in meaningful, efficacious, and ethical ways. In this first phase of the study we explored the question What do young people understand about data within the context of their everyday lives and in relation to personal data management. We present here the findings from a series of semi‐structured interviews with young people, ages 11 to 18, that examined teens perceptions and general knowledge of data in their lives. Results suggest that the teens in this study had varying interpretations of the nature of data and a broad understanding of the lifecycle of data, but most found it difficult to connect with data at a concrete and personal level, with the notion of a personal data dossier either non‐existent or abstract.

What Sources to Rely on:: Laypeople's Source Selection in Online Health Information Seeking

In this study, we examined what sources laypeople would select (i.e., visit and adopt) to resolve their health-related information needs, and how different health conditions affect the selection. Twenty-four college students participated in this user study, where they were asked to search for two separate health issues respectively: multiple sclerosis and weight loss. The search logs were collected and analyzed afterwards. We classify the online information sources on both website level and webpage level, and a webpage classification scheme based on genre is proposed. Results suggest that users» selection of sources depends on different types of health issues in terms of urgency and complexity. Health-specific webpage is a popular source and highly adopted for both tasks, but it is particularly helpful for urgent and complex health conditions. Search engines could facilitate users to navigate among scattered health information and support concerns regarding common health issues.

Affective, Behavioral, and Cognitive Aspects of Teen Perspectives on Personal Data in Social Media: A Model of Youth Data Literacy

13th International Conference on Transforming Digital Worlds, iConference 2018; Sheffield; United Kingdom; 25 March 2018 到 28 March 2018

Social science data repositories in data deluge: A case study of ICPSR’s workflow and practices

Purpose n n n n nOwing to the recent surge of interest in the age of the data deluge, the importance of researching data infrastructures is increasing. The open archival information system (OAIS) model has been widely adopted as a framework for creating and maintaining digital repositories. Considering that OAIS is a reference model that requires customization for actual practice, this paper aims to examine how the current practices in a data repository map to the OAIS environment and functional components. n n n n nDesign/methodology/approach n n n n nThe authors conducted two focus-group sessions and one individual interview with eight employees at the world’s largest social science data repository, the Interuniversity Consortium for Political and Social Research (ICPSR). By examining their current actions (activities regarding their work responsibilities) and IT practices, they studied the barriers and challenges of archiving and curating qualitative data at ICPSR. n n n n nFindings n n n n nThe authors observed that the OAIS model is robust and reliable in actual service processes for data curation and data archives. In addition, a data repository’s workflow resembles digital archives or even digital libraries. On the other hand, they find that the cost of preventing disclosure risk and a lack of agreement on the standards of text data files are the most apparent obstacles for data curation professionals to handle qualitative data; the maturation of data metrics seems to be a promising solution to several challenges in social science data sharing. n n n n nOriginality/value n n n n nThe authors evaluated the gap between a research data repository’s current practices and the adoption of the OAIS model. They also identified answers to questions such as how current technological infrastructure in a leading data repository such as ICPSR supports their daily operations, what the ideal technologies in those data repositories would be and the associated challenges that accompany these ideal technologies. Most importantly, they helped to prioritize challenges and barriers from the data curator’s perspective and to contribute implications of data sharing and reuse in social sciences.

Towards an Integrated Clickstream Data Analysis Framework for Understanding Web Users’ Information Behavior

Clickstream data offers an unobtrusive data source for understanding web users’ information behavior beyond searching. However, it remains underutilized due to the lack of structured analysis procedures. This paper provides an integrated framework for information scientists to employ in their exploitation of clickstream data, which could contribute to more comprehensive research on users’information behavior. Our proposed framework consists of two major components, i.e., data preparation and data investigation. Data preparation is the process of collecting, cleaning, parsing, and coding data, whereas data investigation includes examining data at three different granularity levels, namely, footprint, movement, and pathway. To clearly present our data analysis process with the analysis framework, we draw examples from an empirical analysis of clickstream data of OPAC users’ behavior. Overall, this integrated analysis framework is designed to be independent of any specific research settings so that it can be easily adopted by future researchers for their own clickstream datasets and research questions.

Automatic classification of citation function by new linguistic features

Citation function presents the functional role of a reference in its citing article. These functional information enrich the citation analysis in a semantic perspective and can be used for improving the applications of citation analysis. Though many works on automatic classification have been done, the performance of existing studies cannot satisfy the requirement of analysis on large-scale academic data. In order to overcome the performance bottleneck, in this poster we present some useful features by analyzing and finding unique linguistic patterns in citation context. Our experiments on existing dataset shows the effectiveness of these new features with Support Vector Machine. The performance reaches 86.54% accuracy and a macro F-score of 0.795, which gains an improvement over 20% than previous study on the same dataset.

Types of Tags for Annotating Academic Blogs

Academic blog sites are popular academic information exchange platforms, and they have been widely used in recent years. Blogs in those sites are often annotated with tags, and the tags can help to describe, organize and retrieve these blogs. However, it is still unknown what types of tags are frequently adopted for annotating academic blogs. In this poster, we present survey results for detecting the usage of tag types, and its changes with the bloggers’ demographic information. We believe that our study can benefit users in their access to academic blogs and help the academic blog websites improve their services.

Automatic Course Website Discovery from Search Engine Results

With the rapid development of Internet Technology, the forms of education have been undergoing drastic changes. Instructors are used to posting teaching materials on course websites and setting them publicly accessible. Thus large amounts of course resources have been well organized and shared, which also provide possibilities for building knowledge graphs for a specific domain. However, so far no specific method has been developed for collecting online course resources. In this paper, we propose a method to identify course websites by filtering search results from a general search engine. Experiment results show that the proposed method could achieve good performances on both within-domain and cross-domain tasks, which lays a solid foundation for further work on mining and integrating the online educational resources.