简单说：
弹幕发送者ID是他mid的CRC32.
原文如下。

 

https://github.com/fuckbilibili/Danmaku-De-annoymous/blob/master/README.md

Danmaku-De-annoymous

A simple webpage to dig out some Keyboard Man

Fancy, but how does it works?

Let me try to put this in human language.
Every article of Bilibili would be assigned an unique aid: this is the so-called "av number".
There can be 1 or more pages under one aid. For every content(most of the times are videos) in this page, a unique content ID, cid is assigned.
By calling the view API, we can retrieve the cid with aid: should there exists more pages, a page should be attached to specific one particular content.
The hash is generated as: (all the example code comes from Biligrab)

def calc_sign(string):
    """str/any->str
    return MD5."""
    return str(hashlib.md5(str(string).encode('utf-8')).hexdigest())

The complete API call looks like:

str2Hash = 'appkey={APPKEY}&id={vid}&page={p}&type=xml{SECRETKEY}'.format(APPKEY = APPKEY, vid = vid, p = p, SECRETKEY = SECRETKEY)
biliurl = 'https://api.bilibili.com/view?appkey={APPKEY}&id={vid}&page={p}&type=xml&sign={sign}'.format(APPKEY = APPKEY, vid = vid, SECRETKEY = SECRETKEY, p = p, sign = calc_sign(str2Hash))

The sample response is:

<?xml version="1.0" encoding="UTF-8"?>
<info>
    Blahblah...
  <cid>4802847</cid>
  blahblah
</info>

The view API is also available with JSONP output.
The danmaku file is available at http://comment.bilibili.com/{cid}.xml. In this case I do not count historical comments.
The format of a particular line of danmaku looks like:

<d p="12.456999778748,1,25,16777215,1444811244,0,550e9706,1278188533">第二</d>

in human language:

<d p="time,mode,fontsize,colour,timestamp,danmaku-pool,CRC32-of-User-ID,danmaku-ID">content</d>

It is possible to run a brute-force decrypt of CRC32; or, the preferred way is a small rainbow table with cache, which would significantly reduce the server load.
The user ID is called mid.
The user space is available at http://space.bilibili.com/{mid}.
Jackpot.