NiceT: Named Entity Cleaning and Enhancement with Human-in-the-loop

Named Entities (NEs, in short) are frequent in many datasets, e.g., journalistic investigations (people, places, companies), market analysis (companies and officers), etc. NEs often appear under varied forms, due to spelling variants or mistakes. To leverage NE-rich datasets, the NEs need to be clean (error-free), and possibly enriched with information from external sources. Existing data cleaning solutions tools only partially answer the challenges raised by NE cleaning (lack of regular-structured attributes, need for interactive human input) and/or they do not exploit the opportunities it raises (semantic signals). We propose to demonstrate NiceT, a dedicated NE cleaning tool. Through a visual workflow interface , NiceT allows comparing, clustering, and enriching based on Knowledge Bases and LLMs, and also interacting with the NE cleaning process , down to the granularity of an attribute in a record. The latter is crucial in order to capture advanced knowledge that only domain experts have , and which may be absent from all other sources of information (KB, LLM, etc.) For explainability, NiceT gathers how-provenance that traces the ways in which information contributes to clean NEs. We will showcase it on a variety of real-life datasets inspired by our collaborations with journalists.