Merge pull request #177 from mailgun/obukhov-sergey-patch-1

Update Readme with how to retrain on your own data

README.rst

@@ -129,6 +129,22 @@ start using it for talon.
 .. _EDRM: http://www.edrm.net/resources/data-sets/edrm-enron-email-data-set
 .. _forge: https://github.com/mailgun/forge
 
+Training on your dataset
+------------------------
+
+talon comes with a pre-processed dataset and a pre-trained classifier. To retrain the classifier on your own dataset of raw emails, structure and annotate them in the same way the `forge`_ project does. Then do:
+
+.. code:: python
+
+    from talon.signature.learning.dataset import build_extraction_dataset
+    from talon.signature.learning import classifier as c 
+    
+    build_extraction_dataset("/path/to/your/P/folder", "/path/to/talon/signature/data/train.data")
+    c.train(c.init(), "/path/to/talon/signature/data/train.data", "/path/to/talon/signature/data/classifier")
+
+Note that for signature extraction you need just the folder with the positive samples with annotated signature lines (P folder).
+
+.. _forge: https://github.com/mailgun/forge
 
 Research
 --------

setup.py

@@ -29,7 +29,7 @@ class InstallCommand(install):
 
 
 setup(name='talon',
-      version='1.4.0',
+      version='1.4.5',
       description=("Mailgun library "
                    "to extract message quotations and signatures."),
       long_description=open("README.rst").read(),

talon/html_quotations.py

@@ -94,6 +94,12 @@ def cut_microsoft_quote(html_message):
         #outlook 2007, 2010 (american)
         "//div[@style='border:none;border-top:solid #B5C4DF 1.0pt;"
         "padding:3.0pt 0in 0in 0in']|"
+        #outlook 2013 (international)
+        "//div[@style='border:none;border-top:solid #E1E1E1 1.0pt;"
+        "padding:3.0pt 0cm 0cm 0cm']|"
+        #outlook 2013 (american)
+        "//div[@style='border:none;border-top:solid #E1E1E1 1.0pt;"
+        "padding:3.0pt 0in 0in 0in']|"
         #windows mail
         "//div[@style='padding-top: 5px; "
         "border-top-color: rgb(229, 229, 229); "

talon/quotations.py

@@ -38,10 +38,14 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'Op',
             # German
             'Am',
+            # Portuguese
+            'Em',
             # Norwegian
             u'På',
             # Swedish, Danish
             'Den',
+            # Vietnamese
+            u'Vào',
         )),
         # Date and sender separator
         u'|'.join((
@@ -62,8 +66,12 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'schreef','verzond','geschreven',
             # German
             'schrieb',
+            # Portuguese
+            'escreveu',
             # Norwegian, Swedish
             'skrev',
+            # Vietnamese
+            u'đã viết',
         ))
     ))
 # Special case for languages where text is translated like this: 'on {date} wrote {somebody}:'
@@ -143,7 +151,7 @@ RE_FROM_COLON_OR_DATE_COLON = re.compile(u'(_+\r?\n)?[\s]*(:?[*]?{})[\s]?:[*]?.*
 RE_ANDROID_WROTE = re.compile(u'[\s]*[-]+.*({})[ ]*[-]+'.format(
     u'|'.join((
         # English
-        'wrote'
+        'wrote',
     ))), re.I)
 
 # Support polymail.io reply format
@@ -161,15 +169,15 @@ SPLITTER_PATTERNS = [
     RE_FROM_COLON_OR_DATE_COLON,
     # 02.04.2012 14:20 пользователь "bob@example.com" <
     # bob@xxx.mailgun.org> написал:
-    re.compile("(\d+/\d+/\d+|\d+\.\d+\.\d+).*@", re.S),
+    re.compile("(\d+/\d+/\d+|\d+\.\d+\.\d+).*\s\S+@\S+", re.S),
     # 2014-10-17 11:28 GMT+03:00 Bob <
     # bob@example.com>:
-    re.compile("\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+GMT.*@", re.S),
+    re.compile("\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+GMT.*\s\S+@\S+", re.S),
     # Thu, 26 Jun 2014 14:00:51 +0400 Bob <bob@example.com>:
     re.compile('\S{3,10}, \d\d? \S{3,10} 20\d\d,? \d\d?:\d\d(:\d\d)?'
                '( \S+){3,6}@\S+:'),
     # Sent from Samsung MobileName <address@example.com> wrote:
-    re.compile('Sent from Samsung .*@.*> wrote'),
+    re.compile('Sent from Samsung.* \S+@\S+> wrote'),
     RE_ANDROID_WROTE,
     RE_POLYMAIL
     ]
@@ -282,7 +290,7 @@ def process_marked_lines(lines, markers, return_flags=[False, -1, -1]):
     # inlined reply
     # use lookbehind assertions to find overlapping entries e.g. for 'mtmtm'
     # both 't' entries should be found
-    for inline_reply in re.finditer('(?<=m)e*((?:t+e*)+)m', markers):
+    for inline_reply in re.finditer('(?<=m)e*(t[te]*)m', markers):
         # long links could break sequence of quotation lines but they shouldn't
         # be considered an inline reply
         links = (
@@ -426,6 +434,9 @@ def _extract_from_html(msg_body):
     Extract not quoted message from provided html message body
     using tags and plain text algorithm.
 
+    Cut out first some encoding html tags such as xml and doctype
+    for avoiding conflict with unicode decoding
+
     Cut out the 'blockquote', 'gmail_quote' tags.
     Cut Microsoft quotations.
 
@@ -441,6 +452,9 @@ def _extract_from_html(msg_body):
         return msg_body
 
     msg_body = msg_body.replace(b'\r\n', b'\n')
+
+    msg_body = re.sub(r"\<\?xml.+\?\>|\<\!DOCTYPE.+]\>", "", msg_body)
+
     html_tree = html_document_fromstring(msg_body)
 
     if html_tree is None:

talon/signature/extraction.py

@@ -32,7 +32,7 @@ RE_REVERSE_SIGNATURE = re.compile(r'''
 
 def is_signature_line(line, sender, classifier):
     '''Checks if the line belongs to signature. Returns True or False.'''
-    data = numpy.array(build_pattern(line, features(sender)))
+    data = numpy.array(build_pattern(line, features(sender))).reshape(1, -1)
     return classifier.predict(data) > 0
 
 

tests/text_quotations_test.py

@@ -119,6 +119,38 @@ On 11-Apr-2011, at 6:54 PM, Roman Tkachenko <romant@example.com> sent:
     eq_("Test reply", quotations.extract_from_plain(msg_body))
 
 
+def test_appointment():
+    msg_body = """Response
+
+10/19/2017 @ 9:30 am for physical therapy
+Bla
+1517 4th Avenue Ste 300
+London CA 19129, 555-421-6780
+
+John Doe, FCLS
+Mailgun Inc
+555-941-0697
+
+From: from@example.com [mailto:from@example.com]
+Sent: Wednesday, October 18, 2017 2:05 PM
+To: John Doer - SIU <jd@example.com>
+Subject: RE: Claim # 5551188-1
+
+Text"""
+
+    expected = """Response
+
+10/19/2017 @ 9:30 am for physical therapy
+Bla
+1517 4th Avenue Ste 300
+London CA 19129, 555-421-6780
+
+John Doe, FCLS
+Mailgun Inc
+555-941-0697"""
+    eq_(expected, quotations.extract_from_plain(msg_body))
+
+
 def test_line_starts_with_on():
     msg_body = """Blah-blah-blah
 On blah-blah-blah"""
@@ -401,6 +433,14 @@ Op 17-feb.-2015, om 13:18 heeft Julius Caesar <pantheon@rome.com> het volgende g
 Small batch beard laboris tempor, non listicle hella Tumblr heirloom.
 """))
 
+def test_vietnamese_from_block():
+    eq_('Hello', quotations.extract_from_plain(
+    u"""Hello
+
+Vào 14:24 8 tháng 6, 2017, Hùng Nguyễn <hungnguyen@xxx.com> đã viết:
+
+> Xin chào
+"""))
 
 def test_quotation_marker_false_positive():
     msg_body = """Visit us now for assistance...
@@ -770,3 +810,16 @@ def test_split_email():
     expected_markers = "stttttsttttetesetesmmmmmmssmmmmmmsmmmmmmmm"
     markers = quotations.split_emails(msg)
     eq_(markers, expected_markers)
+
+
+
+def test_feedback_below_left_unparsed():
+    msg_body = """Please enter your feedback below. Thank you.
+
+------------------------------------- Enter Feedback Below -------------------------------------
+
+The user experience was unparallelled. Please continue production. I'm sending payment to ensure
+that this line is intact."""
+
+    parsed = quotations.extract_from_plain(msg_body)
+    eq_(msg_body, parsed.decode('utf8'))