As part of the EYRE '19 workshop co-located with CIKM 2019, two shared tasks related to entity retrieval are organized, in order to assess strengths and weaknesses of existing systems, compare performance of techniques, and enhance communication among researchers and developers.
An entity in a knowledge base is involved in a set of triple-structured facts that describe the entity. When there are many facts, entity summarization is the task of selecting a size-constrained subset of triples that best represent the entity.
For this shared task, we use the Entity Summarization BenchMark v1.1. The benchmark has provided a Java tool for evaluating system-generated entity summaries based on a crowdsourced gold standard. Please note that the benchmark is based on two datasets: DBpedia and LinkedMDB. Participants can report their results (i.e., F-measure under k=5 and k=10) for either or both datasets, but are encouraged to report both.
Entity search is the task of returning a ranked list of entities in a knowledge base in response to a free text user query.
For this shared task, we use the DBpedia-Entity v2 benchmark. The benchmark has provided a crowdsourced gold standard for evaluating system-generated ranked entities. The provided train/test split should be followed. Please note that the benchmark categorizes queries into four groups: named entity queries, IR-style keyword queries, natural language questions, and queries that seek a particular list of entities. Participants can report their results (i.e., NDCG@10 and NDCG@100) for any subset of categories, but are encouraged to report all. NDCG should be calculated using trec_eval.
Each participating system should submit a short paper (2-4 pages) describing the system and reporting the results. Participants should publish the outputs and source codes of their systems, and include a link to those resources in their papers.
Manuscripts should be submitted electronically, as PDF formatted using the ACM camera-ready templates; please use sample-sigconf.tex as the template. Papers should be submitted through the EYRE '19 online submission system.
Papers will be reviewed by the organizers. Selected papers will be included in the workshop proceedings. At least one author of each selected paper is expected to attend the workshop and present.