It would be nice to also rank Wikidata lexemes. Some of the ranking signals would be the same as with Wikidata entities, since Wikidata lexemes are also regular Wiki pages.
However, to be actually useful, we’d need to consider lexical frequency. Perhaps computing word counts on rendered (template-expanded) Wikipedia pages, with a decent word segmentation for languages that don’t make word boundaries explicit, possibly combining with word frequencies from Corpus Crawler. We’d need to decide how to split frequencies to homographs. Overall, this task sounds big enough to justify a separate project.
It would be nice to also rank Wikidata lexemes. Some of the ranking signals would be the same as with Wikidata entities, since Wikidata lexemes are also regular Wiki pages.
However, to be actually useful, we’d need to consider lexical frequency. Perhaps computing word counts on rendered (template-expanded) Wikipedia pages, with a decent word segmentation for languages that don’t make word boundaries explicit, possibly combining with word frequencies from Corpus Crawler. We’d need to decide how to split frequencies to homographs. Overall, this task sounds big enough to justify a separate project.