Wall Observer Observer - MtGoxUSD wall movement tracker tracker

[bb113]

Yes, also, anyone want to help on the regex to extract only the non-quote text from this:

Code:

							<div class="post"><div class="quoteheader"><a href="https://bitcointalk.org/index.php?topic=85687.msg945087#msg945087">Quote from: proudhon on June 06, 2012, 05:38:47 PM</a></div><div class="quote"><div class="quoteheader"><a href="https://bitcointalk.org/index.php?topic=85687.msg945072#msg945072">Quote from: Piper67 on June 06, 2012, 05:31:01 PM</a></div><div class="quote"><div class="quoteheader"><a href="https://bitcointalk.org/index.php?topic=85687.msg945048#msg945048">Quote from: Spekulatius on June 06, 2012, 05:19:16 PM</a></div><div class="quote">a corresponding timesstamp would be nice<br /></div><br />Hmmm... aren&#039;t the messages on the forum time-stamped already?<br /></div><br />Yes, but he posted images that he&#039;d taken hours or days before he started the thread, I believe.<br /></div><br />yes the first 2 do not have a correct times stamp but are relatively new<br /><br />the time from the posts should be fine<div class="quoteheader"><a href="https://bitcointalk.org/index.php?topic=85687.msg945100#msg945100">Quote from: casascius on June 06, 2012, 05:49:37 PM</a></div><div class="quote">I wonder why turning on the &quot;show volume&quot; (the option that darkens the green line anywhere the volume spikes) isn&#039;t simply on by default.&nbsp; It&#039;s extra information at a cost of zero screen clutter.<br /></div><br />thanks i was forgetting about this option</div>

Its this post:

a corresponding timesstamp would be nice

Hmmm... aren't the messages on the forum time-stamped already?

Yes, but he posted images that he'd taken hours or days before he started the thread, I believe.

yes the first 2 do not have a correct times stamp but are relatively new

the time from the posts should be fine

I wonder why turning on the "show volume" (the option that darkens the green line anywhere the volume spikes) isn't simply on by default.  It's extra information at a cost of zero screen clutter.

thanks i was forgetting about this option

[1713982781]

1713982781

[1713982781]

1713982781

[1713982781]

1713982781

[1713982781]

1713982781

[bb113]

I gave up on the sentiment index for now, but here is a new type of plot. The diameter of each "Bead on a VWAP" corresponds to the number of posts by either Proudhon (who has been bearish this whole time if I remember correctly), and Loaded, who seems to post before interesting things occur. Also the volume and posts per day charts are now of the same type.

[bb113]

Ok, so I have important stuff to do but keep procrastinating reading bitcoin stuff so I may as well improve this. I need a list of bullish and bearish terms to create a sentiment index for that thread.

Classify days as bullish/bearsish/neutral.  Use TF-IDF to pick out the most frequent terms on bullish or bearish days.

Do you mean just choose a few as training samples and subjectively classify them, or is there a specific clustering algorithm that is good for this type of thing? It's all new to me.

[notme]

Ok, so I have important stuff to do but keep procrastinating reading bitcoin stuff so I may as well improve this. I need a list of bullish and bearish terms to create a sentiment index for that thread.

Classify days as bullish/bearsish/neutral.  Use TF-IDF to pick out the most frequent terms on bullish or bearish days.

Do you mean just choose a few as training samples and subjectively classify them, or is there a specific clustering algorithm that is good for this type of thing? It's all new to me.

1. Classify time period as bullish/bearish/neutral based on price action

2.  Gather statistics on word frequency per time period as well as word frequencies from a general text corpus.

3.  Use statistics to compute the frequency of each word for each class (bullish, bearish, neutral).  These are your term frequencies for that class.  Weight each term frequency by the inverse of the document frequency (a bit of terminology here: document frequency in this context refers to the frequency of the word in the general text corpus).  This weighting helps filter out common words and accentuate words that are unique to the target class.

4.  Pick the top few TF-IDF words to predict for each class.