Topic saliency is slow right now because it's recomputing per-token constantly, which works for the individual document previews but not for sorting. We should store this information to reuse if we're sorting. (For now, we're factoring this out for the user study.)
Topic saliency is slow right now because it's recomputing per-token constantly, which works for the individual document previews but not for sorting. We should store this information to reuse if we're sorting. (For now, we're factoring this out for the user study.)