Wei

@jannson 之前我也有调查过 term lookup 的数据结构，听说是中文不太适合字典树，状态机省内存但又比较难以理解（至少我看 whoosh的实现的感觉），所以感觉大家做中文搜索的都在用 DAT，当然我没有确切的统计数据。

Would it be OK to just use `vector::iterator` to go through the words, count and assign the "sequence number" you want? If you are using this char offset to evaluate...

TextRank的计算复杂度那么高，想必提出来一定是有效果上的优势的。

@j-dec You are right, but completely ignoring parentheses will result in no difference between `c(x)` and `cx`, the former is a function but the latter is multiplication. I have to...

@GaurangTandon Thank you for reporting, will investigate later when I get some time.

@GaurangTandon Hi, the reason is actually quite simple, since the search result snippet is trying to summarize a document in a short paragraph, it has to skip some content and...

Same question here as what @qiuyuchen14 asked.

@da03 I am wondering how you can replace the horizontal fraction line into hand-written style as shown in your paper, any idea?

@da03 Thank you for your reply, that helps a lot.

The evaluation time using the current colbert code is estimated by tqdm: > 13/6980 [05:46