Ask HN: How does Microsoft calculate commercial intent?

Human evaluators labeled non-monetized query terms from the first week of Dec ’06. The queries represented the top 70% of queries (by frequency… 1000 sampled per 10% frequency bin). Some of the bot-like queries and most non-English language queries were removed to arrive at 6,258 queries. These were labeled as commercial, non-commercial, unsure, or foreign language queries. Of the 6,258 queries, 5,017 were labeled as ‘commercial’ or ‘non-commercial’ by the majority of the evaluators. This analysis looks at these 5K queries.