Cumulative citation recommendation refers to the problem faced by

knowledge base curators, who need to continuously screen the media for

updates regarding the knowledge base entries they manage. Automatic

system support for this entity-centric information processing problem

requires complex pipe\-lines involving both natural language

processing and information retrieval components. The default pipeline

involves four stages: filtering, classification, ranking (or scoring),

and evaluation. Filtering is an initial step, that reduces the

web-scale corpus of news and other relevant information sources that

may contain entity mentions into a working set of documents that should

be more manageable for the subsequent stages.

This step has a large impact on the recall that can be achieved.

Keeping the subsequent steps constant, we therefore zoom in into the

filtering stage, and conduct an in-depth analysis of the main design

decisions here:

cleansing noisy web data, the methods to create entity profiles, the

types of entities of interest, document type, and the grade of

relevance of the document-entity pair under consideration.

We analyze how these factors (and the design choices made in their

corresponding system components) affect filtering performance.

We identify and characterize the relevant documents that do not pass

the filtering stage by examing their contents. This way, we give

estimate a practical upper-bound of recall for entity-centric stream

filtering.