Export
Research Overview
Jianmin Zhang focuses on natural language processing, specifically automated text summarization, news synthesis, and the extraction of reliable information from unstructured text. Early work explored incorporating information certainty into summarization systems, while more recent research investigates the automatic construction of coherent overview articles from diverse news sources and live commentary.
Research Trajectory
Jianmin Zhang’s research trajectory demonstrates a sustained focus on improving the quality, reliability, and coherence of automated text summarization, particularly within the domain of news generation. The early phase of this work addressed the qualitative nature of the information being extracted. In 2014, Zhang co-authored CTSUM: extracting more certain summaries for news articles. Recognizing that standard summarization systems often fail to distinguish between verified facts and speculative statements, this research introduced a system to predict the certainty levels of sentences using support vector regression. By incorporating information certainty as a core factor, the work aimed to provide readers with more reliable news summaries.
By 2016, Zhang’s focus expanded from single-document extraction to the synthesis of narratives from dynamic, real-time data sources. As first author, Zhang published Towards Constructing Sports News from Live Text Commentary. This work tackled the challenge of transforming fragmented, chronological updates into structured news reports. This approach aligns closely with the research direction Coherent Synthesis from Extractive Text Fragments, which seeks to avoid the disjointed "Frankenstein summaries" often produced by purely extractive methods. Furthermore, processing live commentary requires handling informal and irregular text, connecting this research to the broader area of Robust Natural Language Processing for Noisy Social Media Data.
In 2017, Zhang advanced these synthesis techniques to handle multiple formal news sources. Leading the research in Towards Automatic Construction of News Overview Articles by News Synthesis, Zhang proposed a comprehensive pipeline involving passage segmentation, ranking, selection, and merging. This methodology was designed to automatically construct a cohesive overview article from a disparate set of reports about a single news event. This work directly addresses challenges in the area of Managing Information Overlap in Multi-Document Summarization, where the primary bottleneck is filtering redundant information across sources. The necessity of selecting the most relevant and diverse passages also touches upon methodologies found in the research direction Determinantal Point Processes for Diverse Subset Selection.
Current Directions
Based on the most recent papers in the evidence, which date to 2017, Zhang’s latest documented focus is on the architectural challenges of multi-document news synthesis. The transition from extracting individual sentences to merging segmented passages indicates a growing emphasis on document-level coherence and narrative flow.
This recent work investigates how to maintain logical progression when assembling text from multiple authors and sources. By prioritizing passage segmentation and merging, Zhang's research engages with themes such as Decoupling Content Planning from Surface Realization and the broader area of Sentence Constrained Topic Modeling. The overarching goal remains the automated generation of highly readable, comprehensive, and factually grounded overview articles that rival human-authored summaries.
Profile updated
Loading new papers…
Loading research neighbors…
See something wrong? Report a mistake