PubMed related articles: a probabilistic topic-based model for content similarity

Table 4 Values of pmra parameters (λ, μ) estimated using different sets of MEDLINE citations.

Set Used	Size	λ	μ	P5
All assessed documents from TREC 2005 genomics track	39874	0.032	0.022	0.397°
Top 100 hits for every relevant citation, bm25	453402	0.023	0.013	0.398°
Top 100 hits for every template query, Indri	4991	0.022	0.012	0.397°
Top 1000 hits for every template query, Indri	49907	0.024	0.013	0.397°
Optimal parameters		0.022	0.013	0.399

We see that estimated values are close to optimal parameter values in many cases, and that differences in P5 performance are not statistically significant.

ISSN: 1471-2105